4.3. Digitized Rollei 6006 images
For the comparison of the results derived with the digital
photogrammetric system and a conventional system un-
der similar conditions, a stereo pair of the front facade
was digitized with the scanner Optronics 5040. The.
6 x 6 cm? colour slides of the Rollei 6006 were enlarged
to 24 x 24 cm? colour prints. This was necessary, because
the used scanner accepts only opaque media. The print
was scanned with a resolution of 50 um. This corre-
sponds to scanning the slide with a 12.5 um resolution.
The resulting digital image has 5000 x 5000 pixel, which
is comparable to the resolution of a medium format film-
based camera.
The difference in resolution between the Rollei 6006 and
the Canon CI-10 is demonstrated in figure 6. It shows
zoomed parts of the front facade. The big difference of
these two imaging systems is conspicuous. This demon-
strates the need for high-resolution solid-state imaging
systems for architectural photogrammetry.
5. MEASUREMENT TECHNIQUES IN
DIGITAL ARCHITECTURAL
PHOTOGRAMMETRY
A system for digital photogrammetry offers various
methods for the measurement of image coordinates.
When using semi-automatic methods, the operators task
is only to judge the scene qualitatively, whereas the quan-
titative statement (measurement) is done by computer.
The result is not affected by the subjective human meas-
urement. Furthermore the operator is supported in such a
way, that all measurements and other known informa-
tions are visualized on the screen. This is comparable
with the super-imposition technique used by analytical
plotters. Double measurements or confusion of points are
reduced to a minimum.
Three different measurement techniques are currently im-
plemented in DEDIP: manual point location (chapter
5.1.), point location using Least Squares Template
Matching (chapter 5.2.), and feature location via line
tracking (chapter 5.3.).
5.1. Manual point location
The simplest method to determine the location of archi-
tectural features in the images is to use the cursor as a
manual measuring device. This is basically identical to
Patch at initial
Part of image with
signalized point position
the measurement technique used in many systems em-
ploying a digitizer (e.g. Rolleimetric MR2, Elcovision
10). In DEDIP the regions of the images, in which the co-
ordinates have to be determined, are displayed on the
screen and the operator measures the coordinates of cor-
responding points with the cursor. The images are usually
zoomed to improve the precision of the manual measure-
ments. Several regions of interest can be viewed simulta-
neously. The measured coordinates are indicated with
crosses and their respective point numbers. This meas-
urement technique does not exploit the additional capa-
bilities offered by semi-automatic measurement methods.
5.2. Point location using Least Squares Template
Matching
In this project Least Squares Template Matching (LSTM,
Gruen, 1985) was applied to measure the precise position
of signalized points. It uses a template (artificial image of
the signalized point) as reference and determines the po-
sition via an iterative procedure through the least squares
fit of an affine transformation between template and patch
(image region). Initially, the patch is taken at the approxi-
mate position from the image. In general this can be indi-
cated by the operator in an interactive mode. In
subsequent iterations the patch is resampled from the data
of the image using updated values for the affine transfor-
mation from an interpolation algorithm. Figure 7 shows
an enlargement of a part of the original image, the patch
at the initial position, the patch after convergence of the
algorithm, and the template. Point location using LSTM
provides a very high precision. In practical applications
an accuracy corresponding to 1/10” of the pixel spacing
can be achieved. Under laboratory conditions accuracies
of a few hundreds of a pixel have been achieved.
5.3. Feature location via line tracking
Looking at images for architectural photogrammetry one
can see that in the most cases linear boundaries of an ar-
chitectural feature contain more information than the ver-
tices of this feature. The measurement technique shown
here, takes advantage of this knowledge. It first locates
the linear elements of the feature to be measured and then
derives the vertices as intersections of these lines.
The geometric and semantic information of the object is
thus generated together during the measurement process,
which is important for architectural processing. Beside a
list of three-dimensional object coordinates a surface
Patch at final Template
position
Figure 7: Image with point and number at final position, patches at initial and final position, and template of
a target measured with Least Squares Template Matching.
p
uud — (4 — von — 0/7 end bude FAN Ar FA Peed may PA mp 635; — 2 OPN gw bed
i -— i Aaa Ln | Re ouf paquet << 2e 71 nat
- uc d suo <a ultus
